Generative AI is able to create software code due to it's understanding of language and its ability to predict a reasonable chain of tokens based on the probability weightings and alignment mechanisms of LLM's. In other words, software codeis just another form of language like English, French and the like. At times this appears magical, as it can generate quite a bit of code with just a single prompt. However, there are many aspects of full-scale software architecting where LLM's by themselves are insufficient. To achieve software architecture agency to autonomously generate full solutions requires attention to such issues as planning, automation and inference. These capabilities must be "declared" from knowledge that augments the capabilities of the stochastic LLM engine. Such Declarative AI has been around for a long time in the form of "Expert Systems".
Working in concert with the generative capacity of LLM's, an Expert System codify the guard-rails by which solutions can emerge through the iteratively converging strategies of long-term planning automation that drives the prompting. It persists the types of prompts necessary to construct the various layers of modern software architectures to yield observability in what is being generated. It provides choice and coherence of technology stacks such that technical debt and obsolescence are mitigated. It moves the software foundry capabilities leveraging LLM's from black-box to white-box.
We have been very busy these past few years incubating these necessary capabilities to yield repeatable enterprise value out of LLM's. We call this capability Advisor™ :: Architect.
​
